Global music body urges Australia to reject weaker copyright rules for AI training

Australia has promised strong copyright protection for creators. Now Victoria Oakley, head of the global recording-industry body IFPI, is warning that concessions for AI training would undermine that promise. In an interview with Guardian Australia, she called weaker protections a “disaster for creativity” whose consequences could extend well beyond Australian music.

The Albanese government is considering copyright changes as part of its AI policy work; it has not announced a change. AI companies want greater legal certainty before investing in Australia or training models there. Oakley disputes the premise that attracting that investment requires giving developers the copyright terms they seek.

One country’s rules, a worldwide concern

IFPI represents recording-industry bodies worldwide. Oakley’s concern is not limited to whether Australian artists receive payment: she argues that a permissive jurisdiction could become a place to train on music originating anywhere. She also fears other governments could follow Australia in relaxing protections.

She said IFPI had been raising Australia’s deliberations with governments elsewhere. Her warning includes reputational damage to Australia within the creative industries. She declined to speculate about possible legal or constitutional challenges, keeping her argument focused on the policy choice and its consequences for music rights holders.

I just don’t think it’s true that you will only get investment or growth, or be seen as pro-innovation, if you entirely give in to the demands of AI companies.

Victoria Oakley, IFPI CEO, speaking to Guardian Australia

What an opt-out would permit

Anthropic has proposed allowing content scraping unless rights holders object. Separately, ABC reported on September 15 that government consultation documents contained two options making inclusion in AI training opt-out by default. Those were consultation proposals, not adopted rules. They show how officials have explored access to online material beyond individually negotiated licences.

• One option included payments to a central body that would distribute money to registered copyright owners. It also offered a route through deals with collecting groups, potentially covering work owned by people outside those groups.
• Another would require developers to make deals with a minimum number of companies for a minimum period. They could then train on material whose owners had not opted out, without additional payments beyond the qualifying deals.

The documents described the problem as the “long tail”: online material for which voluntary deals would be unrealistic or impossible. Proposed safeguards included penalties for ignoring opt-outs, transparency and audit requirements, protections for Indigenous cultural and intellectual property, and “best efforts” to avoid training on pirated material.

The dispute over licensing

Oakley counters that music labels have already signed dozens of licensing agreements with smaller AI companies. She presents those deals as evidence that licensing is workable, rather than an obstacle that requires weaker copyright law. Rights holders, she said, are willing to negotiate with developers.

OpenAI has challenged the comparison with music streaming. Answering parliamentary questions about why it could not replicate Spotify’s deals, it said Spotify primarily transmits “un-transformed copyrighted content” and would therefore be expected to license what it streams. Its answer distinguishes streaming from model training; Oakley points instead to agreements already being made within AI itself.

The government continues to promise protection. Attorney general Michelle Rowland’s spokesperson said last week that any future copyright changes would deliver “meaningful control and fair compensation”. Prime minister Anthony Albanese has promised “the strongest possible protection” and said failing to pay creatives for their work would be “theft”.